Skip to content

Convert BART models to ONNX with quantization. 3X reduction in size, and upto 3X boost in inference speed

Notifications You must be signed in to change notification settings

siddharth-sharma7/fast-Bart

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

4 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

fast-Bart

Reduction of BART model size by 3X, and boost in inference speed up to 3X

BART implementation of the fastT5 library (https://github.com/Ki6an/fastT5)

Pytorch model -> ONNX model -> Quantized ONNX model


Install

Install using requirements.txt file

git clone https://github.com/siddharth-sharma7/fast-Bart
cd fast-Bart
pip install -r requirements.txt

Usage

The export_and_get_onnx_model() method exports the given pretrained Bart model to onnx, quantizes it and runs it on the onnxruntime with default settings. The returned model from this method supports the generate() method of huggingface.

If you don't wish to quantize the model then use quantized=False in the method.

from fastBart import export_and_get_onnx_model
from transformers import AutoTokenizer

model_name = 'facebook/bart-base'
model = export_and_get_onnx_model(model_name)

tokenizer = AutoTokenizer.from_pretrained(model_name)
input = "This is a very long sentence and needs to be summarized."
token = tokenizer(input, return_tensors='pt')

tokens = model.generate(input_ids=token['input_ids'],
               attention_mask=token['attention_mask'],
               num_beams=3)

output = tokenizer.decode(tokens.squeeze(), skip_special_tokens=True)
print(output)

to run the already exported model use get_onnx_model()

you can customize the whole pipeline as shown in the below code example:

from fastBart import (OnnxBart, get_onnx_runtime_sessions,
                    generate_onnx_representation, quantize)
from transformers import AutoTokenizer

model_or_model_path = 'facebook/bart-base'

# Step 1. convert huggingfaces bart model to onnx
onnx_model_paths = generate_onnx_representation(model_or_model_path)

# Step 2. (recommended) quantize the converted model for fast inference and to reduce model size.
# The process is slow for the decoder and init-decoder onnx files (can take up to 15 mins)
quant_model_paths = quantize(onnx_model_paths)

# step 3. setup onnx runtime
model_sessions = get_onnx_runtime_sessions(quant_model_paths)

# step 4. get the onnx model
model = OnnxBart(model_or_model_path, model_sessions)

                      ...
custom output paths

By default, fastBart creates a models-bart folder in the current directory and stores all the models. You can provide a custom path for a folder to store the exported models. And to run already exported models that are stored in a custom folder path: use get_onnx_model(onnx_models_path="/path/to/custom/folder/")

from fastBart import export_and_get_onnx_model, get_onnx_model

model_name = "facebook/bart-base"
custom_output_path = "/path/to/custom/folder/"

# 1. stores models to custom_output_path
model = export_and_get_onnx_model(model_name, custom_output_path)

# 2. run already exported models that are stored in custom path
# model = get_onnx_model(model_name, custom_output_path)

Functionalities

  • Export any pretrained Bart model to ONNX easily.
  • The exported model supports beam search and greedy search and more via generate() method.
  • Reduce the model size by 3X using quantization.
  • Up to 3X speedup compared to PyTorch execution for greedy search and 2-3X for beam search.

About

Convert BART models to ONNX with quantization. 3X reduction in size, and upto 3X boost in inference speed

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published